Framing and Understanding the Importance of Good Teamwork in Cooperative Settings

In the field of artificial intelligence, researchers are exploring how to train agents to work together effectively in a cooperative setting. This involves understanding what makes a good teammate and developing a curriculum of teammates to train an agent to be helpful and capable in a variety of scenarios. The article discusses the challenges of defining a good teammate, as it depends on the specific context and objectives of the cooperation. The authors propose using a curriculum of teammates to train agents that can adapt to different situations and contribute to the shared task. The experiments in the study show that training an agent with a curriculum of teammates is helpful in achieving the objectives of cooperation, such as maximizing rewards or helping a novice agent learn. The findings suggest that optimizing for these factors can result in agents that are generally capable and adaptable in different cooperative settings.